Red Hot Cyber
Cybersecurity, Cybercrime News and Vulnerability Analysis
With the boom in artificial intelligence, bold claims about breakthroughs in cybersecurity are becoming increasingly common, but not all experts are willing to accept them at face value. A new controversy has erupted around Anthropic’s latest innovation, which the company touts as a significant advancement in vulnerability detection.
British malware analyst Marcus Hutchins , who rose to prominence after stopping the WannaCry outbreak, has publicly questioned Anthropic’s claims about the capabilities of the Mythos model.
The reason was a presentation in which the company claimed that the system was able to find and exploit vulnerabilities better than most humans.
Anthropic also reported that the model has already discovered thousands of previously unknown issues in popular operating systems, browsers, and software. Access to Mythos, however, is currently limited ; it is being tested by leading technology and cybersecurity organizations as part of the Glasswing project.
Hutchins analyzed one of the examples cited: a vulnerability in OpenBSD that the system would have detected with a computational effort of less than $20,000. He estimated that the vulnerability consisted of a null pointer dereference , a type of error that typically causes a system crash rather than a full check. This discovery, according to the expert, does not seem like a major achievement.
The stated price has also raised some questions. Hutchins suggested that the figure reflects the price of API tokens, rather than the actual infrastructure costs. Given the active funding from venture capital funds, the processing cost could be lower.
The specialist identifies the main problem not in technology, but in economics. Vulnerabilities remain unresolved not because of a lack of tools, but because of a lack of motivation and incentives for those who identify them . Even with the use of artificial intelligence, code analysis requires investment, and without a change in the funding model, the situation will not change.
Hutchins also noted that attackers are more likely to resort to social engineering and phishing than sophisticated technical attacks. Therefore, new vulnerability scanning tools don’t fundamentally alter the balance of power in cybersecurity.
The analyst has previously criticized sweeping claims about the role of artificial intelligence in industry. This fall, for example, he disputed an MIT study that found 80% of ransomware attacks were AI-related .
This article was later removed from the university website.
